ModelScan
Protect AI (Palo Alto Networks) · Open Source · Trust 100Scans model files (Pickle, H5, SavedModel) for embedded malicious code without loading the model. By Protect AI, now Palo Alto Networks.
Make it safe · Alternatives
Alternatives to Meta PurpleLlama (Llama Guard 4, Prompt Guard 2, LlamaFirewall)
Meta PurpleLlama (Llama Guard 4, Prompt Guard 2, LlamaFirewall) by Meta belongs to the area “Make it safe” on OROBO. Here are 8 alternatives from the same area, first 8 with the same goals. Every entry is checked automatically every day (origin, code patterns, licence, maintenance, reachability); OROBO does not check whether software fits your purpose.
8 alternatives at a glance
garak
NVIDIA · Open Source · Trust 100Command-line scanner that probes a language model with hundreds of attacks for jailbreaks, leaks and hallucinations.
Guardrails AI
Guardrails AI (Harvey) · Open Source · Trust 100Python framework with validators that check AI inputs and outputs and enforce structured responses. Part of Harvey since September 2026.
NeMo Guardrails
NVIDIA · Open Source · Trust 100Toolkit for programmable guardrails: keeps chatbots on topic, masks personal data and blocks jailbreaks.
Giskard
Giskard · Open Source · Trust 100French open-source tool that scans LLM apps for hallucinations, prompt injection and data disclosure.
PyRIT
Microsoft (AI Red Team) · Open Source · Trust 100Python framework that lets security teams attack generative AI systems automatically to find risks.
promptfoo
Promptfoo (OpenAI) · Open Source · Trust 97Automatically tests prompts, agents and RAG apps and runs red-teaming and vulnerability scans locally. Now part of OpenAI, still open source.
Snyk Agent Scan (formerly Invariant MCP-Scan)
Snyk (ex-Invariant Labs team) · Open Source · Trust 100Scans installed MCP servers and agent skills for tool poisoning, hidden instructions and malware; includes a web inspector.
Keep comparing
All software in “Make it safe” with filters by price, offline use and difficulty.